AI indifference through utility manipulation — LessWrong